Pandas to_sql change column type from varchar to text

Question

I wrote a little script to copy a table between SQL servers. It works, but one of the columns changed type from varchar to text... How do I make it to copy a table with the same columns types?

import pymssql
import pandas as pd
from sqlalchemy import create_engine


db_server= 1.2.3.4\\r2
db_database="Test_DB"
db_user="vaf"
db_password="1234"

local_db_server="1.1.1.1\\r2"
local_db_database="Test_DB"
local_db_user="vaf"
local_db_password="1234"

some_query=("""
select * from some_table
""")




def main():
    conn=pymssql.connect(server=local_db_server,user=local_db_user,password=local_db_password,database=local_db_database,charset='UTF-8')
    data=pd.io.sql.read_sql(some_query,conn)

    connection_string='mssql+pymssql://{}:{}@{}/{}'.format(db_user,db_password,db_server,db_database)
    engine=create_engine(connection_string)
    data.to_sql(name="some_table",con=engine,if_exists='replace',index=False)

if __name__ == "__main__":
    main()

Thanks

Answer 1

Consider three approaches:

SPECIFY TYPES (proactive as it anticipates ahead)

Using the dtype argument of pandas.DataFrame.to_sql, pass a dictionary of sqlalchemy types for named columns.

data.to_sql(name="some_table", con=engine, if_exists='replace', index=False,
            dtype={'datefld': sqlalchemy.DateTime(), 
                   'intfld':  sqlalchemy.types.INTEGER(),
                   'strfld': sqlalchemy.types.VARCHAR(length=255),
                   'floatfld': sqlalchemy.types.Float(precision=3, asdecimal=True),
                   'booleanfld': sqlalchemy.types.Boolean}

DELETE DATA (proactive as it anticipates ahead)

Clean out table with DELETE action query. Then, migrate only the data from pandas to SQL Server without structurally changing table since to_sql replace argument re-creates the table. This approach assumes dataframe is always consistent (no new columns / changed types) with database table.

def main():
   connection_string = 'mssql+pymssql://{}:{}@{}/{}'\
                         .format(db_user,db_password,db_server,db_database)
   engine = create_engine(connection_string)

   # IMPORT DATA INTO DATA FRAME
   data = pd.read_sql(some_query, engine)

   # SQL DELETE (CLEAN OUT TABLE) VIA TRANSACTION
   with engine.begin() as conn:     
      conn.execute("DELETE FROM some_table")

   # MIGRATE DATA INTO DATA FRAME (APPEND NOT REPLACE)
   data.to_sql(name='some_table', con=engine, if_exists='append', index=False)

MODIFY COLUMN (reactive as it fixes ad-hoc)

Alter the column after migration with a DDL SQL statement.

def main():
   connection_string = 'mssql+pymssql://{}:{}@{}/{}'\
                         .format(db_user,db_password,db_server,db_database)
   engine = create_engine(connection_string)

   # IMPORT DATA INTO DATA FRAME
   data = pd.read_sql(some_query, engine)

   # MIGRATE DATA INTO DATA FRAME 
   data.to_sql(name="some_table", con=engine, if_exists='replace', index=False)

   # ALTER COLUMN TYPE (ASSUMING USER HAS RIGHTS/PRIVILEGES)
   with engine.begin() as conn:     
      conn.execute("ALTER TABLE some_table ALTER COLUMN mytextcolumn VARCHAR(255);")

I recommend the second approach as I believe databases should be agnostic to application code like python and pandas. Hence, initial build/re-build of table schema should be a planned, manual process, and no script should structurally change a database on the fly, only interact with data.